Scholarlink

<p>Three-dimensional High-Efficiency Video Coding (3D-HEVC) standard, as the latest 3D video coding standard, has introduced several new coding algorithms for depth map that enables it to have a higher compression ratio compared with previous standards. However, this also leads to a sharp increase in computational complexity, which has become the critical factor to prevent 3D-HEVC from being widely used. To address this problem, a new framework based on Convolutional Neural Networks (CNN) is proposed to achieve fast depth intra-coding in this article. First, we propose an adaptive Quantization Parameter-Attention CNN (QA-CNN) with multiple attention modules to predict the Coding Units (CUs) partition of depth map. After that, according to the relationship between texture map and depth map, we introduce texture map input to form dual-head QA-CNN (DQA-CNN), using texture map CU partition results to correct the depth map results. Finally, we embed DQA-CNN into 3D-HEVC test platform to achieve fast depth intra-coding. Experimental results show that compared with the reference software HTM<InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11554_2025_1684_Article_IEq1.gif" Format="GIF" Height="13" Rendition="HTML" Resolution="72" Type="Linedraw" Width="19" /> </InlineMediaObject> <EquationSource Format="TEX">\(-\)</EquationSource> <EquationSource Format="MATHML"><math> <mo>-</mo> </math></EquationSource> </InlineEquation>16.0, the proposed method can reduce an average of 74.9% of the intra-coding time with the reduction of rate-distortion performance in an acceptable range, outperforming the state-of-the-art methods in reducing coding complexity.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Fast coding based on dual-head attention network in 3D-HEVC

  • 期刊论文

  • 发表日期：

    2025年8月21日

查看全文
  • Yueheng Wu,
  • Kebin Jia

摘要

Three-dimensional High-Efficiency Video Coding (3D-HEVC) standard, as the latest 3D video coding standard, has introduced several new coding algorithms for depth map that enables it to have a higher compression ratio compared with previous standards. However, this also leads to a sharp increase in computational complexity, which has become the critical factor to prevent 3D-HEVC from being widely used. To address this problem, a new framework based on Convolutional Neural Networks (CNN) is proposed to achieve fast depth intra-coding in this article. First, we propose an adaptive Quantization Parameter-Attention CNN (QA-CNN) with multiple attention modules to predict the Coding Units (CUs) partition of depth map. After that, according to the relationship between texture map and depth map, we introduce texture map input to form dual-head QA-CNN (DQA-CNN), using texture map CU partition results to correct the depth map results. Finally, we embed DQA-CNN into 3D-HEVC test platform to achieve fast depth intra-coding. Experimental results show that compared with the reference software HTM \(-\) - 16.0, the proposed method can reduce an average of 74.9% of the intra-coding time with the reduction of rate-distortion performance in an acceptable range, outperforming the state-of-the-art methods in reducing coding complexity.